Three essays in applied macroeconomics and time series analysis
Abi Morshed, Alaa
Publication date: 2017
Document Version
Publisher's PDF, also known as Version of record
Link to publication in Tilburg University Research Portal
Citation for published version (APA):
Abi Morshed, A. (2017). Three essays in applied macroeconomics and time series analysis. CentER, Center for Economic Research.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights. • Users may download and print one copy of any publication from the public portal for the purpose of private study or research. • You may not further distribute the material or use it for any profit-making activity or commercial gain
• You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Series Analysis
PROEFSCHRIFT
ter verkrijging van de graad van doctor aan Tilburg University op gezag van de rector magnificus, prof.dr.E.H.L. Aarts, in het openbaar
te verdedigen ten overstaan van een door het college voor promoties aangewezen commissie in de aula van de Universiteit op dinsdag 30
mei 2017 om 14:00 uur door
ALAA ABI MORSHED
Advance, and never halt, for advancing is perfection. Advance and do not fear the thorns in the path, for they draw only corrupt blood.
Gibran Khalil Gibran First of all, I would like to thank my supervisors, Otilia Boldea and Bas Werker, and the committee members, Massimo Giuliodori, Marie Bri`ere, Peter de Goeij and Bertrand Melenberg for their helpful comments and insights that improved my thesis.
My time in Tilburg would not have been as pleasant without Bas, Mario, Renata and Victor who shared with me good and bad moments, since the start of the research master program almost six years ago. I also would like to thank the following people who I spent with very good time: Evelien, Aida, Marieke, Marleen, Zahra, Steven, Olga, Ahmadreza, Shobeir, Mitzi, Nicola, Roxana, Liz, Bas D., Andreas, Krzysztof, Sebastian, Trevor, Amparo, Nick, Cynthia, Diana, Bo, Yan, Peggy, Emanuel, Inna, Gyula, Ehsan, Maria, Jan and Tina. I really hope that we keep our friendship regardless of the distance. Furthermore, many thanks to Cecile, Korine, Lenie, Anja and Heidi for their support with administrative and job market matters.
Finally, I would like to express my gratitude to my parents, twin sister and brother for their unconditional love and support.
1 Introduction 1
2 Forecast Uncertainty, Disagreement and Financial Markets 3
2.1 Introduction 3
2.2 Data 6
2.2.1 Nominal yields, TIPS, and BEI rates 6 2.2.2 Economic news, disagreement, forecast uncertainty and FUS 6
2.3 Results 11
2.3.1 The sensitivity of Treasury Yields to Macroeconomic Surprises 11 2.3.2 The Sensitivity of Treasury Yields to Disagreement and Forecast
Uncertainty 13
2.3.3 The Sensitivity of Treasury Yields to Macroeconomic Surprises, Forecast Uncertainty and Disagreement 14 2.3.4 The Sensitivity of Treasury Yields to Macroeconomic Surprises and
FUS 16
2.4 Asymmetries in the response of Treasury Yields to Surprises 17
2.4.1 FUS and Squared surprises 19
2.5 Time variation in the response of Treasury yields 19
2.6 Conclusion 21
Appendices 22
Appendix 2.A Tables for Sections 2.3, 2.4 and 2.5 22
Appendix 2.B Robustness checks 43
Appendix 2.C Liquidity adjustment of BEI and TIPS yields 45 3 The Anchoring of Inflation Expectations: A Bayesian Approach 46
3.1 Introduction 46
3.2 News Regressions 49
3.2.1 Constant Parameter News Regressions 49 3.2.2 Time-Varying News-Regression in State Space Representation 50
3.3 Data 51
3.3.1 Market-based measures of inflation expectations 51
3.3.2 Economic News 52
3.4 Bayesian Inference 53
3.4.1 Priors 53
3.8 Macro Determinants of Time Variation 62
3.8.1 Data 62
3.8.2 Results 63
3.9 Conclusion 64
Appendices 65
Appendix 3.A Tables for the results in Chapter 3 65 Appendix 3.B Sensitivity to priors, robustness to alternative specifications and
convergence diagnostics 67
Appendix 3.C Liquidity and Risk adjustment of BEI forward rates 71 Appendix 3.D Plots of the Macroeconomic Determinants 72 Appendix 3.E Robustness to risk and liquidity premia 73 Appendix 3.F Recursive standard deviation in real time 77 Appendix 3.G News regressions with all macroeconomics news 80 Appendix 3.H Sensitivity to the prior of ˆβ 81 Appendix 3.I News effects measured after three days 82 4 Structural break tests robust to regression misspecification 83
4.1 Introduction 83
4.2 Unconditional mean and variance break tests 85 4.3 Conditional mean and variance break tests 88
4.3.1 Correct specification 88 4.3.2 Dynamic Misspecification 90 4.4 Simulation results 92 4.5 Empirical Illustrations 104 4.6 Conclusion 111 Appendices 112
Introduction
This dissertation revolves around topics in Applied Macroeconomics and Time series analysis. Generally speaking, we explore different forms of instability ranging from dis-crete sudden breaks to time varying parameter (TVP) models. In the second chapter, we study the time-varying impact of disagreement and forecast uncertainty about economic fundamentals on nominal yields, treasury-inflation protected securities and market-based inflation expectations. In the third chapter, we employ TVP news regressions to answer a relevant and pressing policy question: whether US long-run inflation expectations (IEs) have become more firmly anchored in the aftermath of the crisis. In the fourth chapter, we enrich the toolkit of policymakers with structural break tests that are robust to re-gression misspecification.
In Chapter 2, we study the impact of second moments (i.e., disagreement and forecast uncertainty) on yields of nominal and inflation-indexed bonds (TIPS) and market-based inflation expectations (IEs) during the period 1999-2016. We use an event-study frame-work and also propose a new measure for forecast uncertainty shocks (FUS). This measure is defined as the difference between root mean squared forecast errors (ex-post forecast uncertainty) and the standard deviation of forecasters’ point forecasts (disagreement). First, we show that most of the responsiveness to second moments is concentrated in nominal yields, while TIPS and IEs react to a bigger number of macroeconomic surprises (i.e., first moment shocks). Second, forecast uncertainty (shocks) about nonfarm payrolls, arguably the most important announcement, affects only nominal yields and its inclusion increases the response of nominal yields by almost one basis point. Third, our results show that Treasury yields react differently to disagreement and forecast uncertainty (shocks) and that the sign of the response depends on the macroeconomic announcement consid-ered, the adjustment of liquidity premia and the macroeconomic regime that prevails. Fourth, we find evidence of asymmetries in the response of Treasury yields to macroeco-nomic announcements, in which negative, positive and large surprises in macroecomacroeco-nomic announcements affect financial markets in different ways. Furthermore, employing struc-tural break tests, we find evidence of time variation in the response of Treasury yields to macroeconomic surprises and second moments.
changes in the communication policy of the Fed and macroeconomic and financial condi-tions. In particular, we find that the explicit inflation target by the Fed in 2012 reduces the response to price and real-side news. Additionally, the state of the labor market and the federal funds rate are important determinants of time variation.
Structural break tests developed in the literature for regression models are sensitive to model misspecification. In Chapter 4 we show - analytically and through simulations - that the sup Wald test for breaks in the conditional mean and variance of a time series process exhibits severe size distortions when the conditional mean dynamics are misspec-ified. We also show that the sup Wald test for breaks in the unconditional mean and variance does not have the same size distortions, yet benefits from similar power to its conditional counterpart. Hence, we propose using it as an alternative and complementary test for breaks. While the conditional tests based on dynamic regression models detect breaks in the mean and variance of the US unemployment growth and interest rate growth series around the Great Moderation, the evidence for these breaks disappears when using the unconditional tests. Therefore, there is no evidence of long-run mean or volatility shifts in unemployment growth and interest rate growth.
Forecast Uncertainty, Disagreement and Financial
Markets
12.1
Introduction
Since the Great Recession, macro uncertainty has been a main concern for policymakers and academics alike. Uncertainty shocks have been shown to be highly countercyclical as they lead to short and sharp drops in output, employment, inflation and the policy rate - see Bloom (2009), Jurado, Ludvigson and Ng (2015), Rossi, Sekhposyan and Soupre (2016), and Bachmann, Elstner and Sims (2013).
Event studies have so far studied the response of asset prices (i.e., bond yields, stocks and exchange rates) to surprises in economic fundamentals, defined as the difference be-tween the announced value of a fundamental and the median forecast – see Balduzzi, Elton and Green (2001), Andersen et al. (2003), Beechey and Jonathan (2009) and Gurkaynak, Sack and Swanson (2005), among others. More recently, Huang (2016) studies the sec-ond moment response of nominal bsec-onds and equity markets to disagreement and market-based, ex-ante uncertainty.2 They find that news surprises and second moments affect
financial market volatility and jump responses, with disagreement and market-based un-certainty having different effects on the latter. Moreover, Pericoli and Veronese (2016) explore whether the response of exchange rates and long term yields to macroeconomic news vary with disagreement across analysts in Bloomberg survey.
This paper attempts to find answers to the following hypotheses: What is the im-pact of disagreement (cross-sectional standard deviation of forecasts) about announced economic fundamentals on financial markets? Does forecast uncertainty (cross-sectional standard deviation of forecast errors) have the same impact as disagreement and what is the sign of this impact? In light of these questions, our paper makes the following contributions to the extant literature. First, we study the first moment response of nom-inal and TIPS yields and market-based IEs to US macroeconomic surprises and second moments from 1999 up to 2016, in an event-study setup. With respect to the related literature, Buraschi and Whelan (2016) put forward a theoretical model with speculative
1This chapter is based on Abi Morshed (2017).
2In a similar fashion, Beber and Brandt (2009) employ a market-based measure of ex-ante uncertainty
of disagreement on nominal, TIPS and IEs but also the impact of forecast uncertainty. Besides that, our setup is different, while we study the impact of second moments about announced macroeconomic fundamentals and their immediate impact on financial mar-kets, their study does not employ this source of identification.5
To our knowledge there is no work done on the impact of forecast uncertainty on the term structure of nominal and TIPS yields and market-based IEs. However, Leippold and Matthys (2015) develop a general equilibrium model in conjunction with an affine yield curve model to study the impact of economic policy uncertainty on the term structure of nominal interest rates. Their empirical analysis suggests that higher government policy uncertainty decreases yields and increases bond yield volatility whereas monetary policy uncertainty does not seem to have any significant impact.6 Regarding the reaction of
survey IEs to economic policy uncertainty, Istrefi and Pilou (2014) employ a SVAR and show that short and long-term IEs rise in response to shocks in economic policy uncer-tainty.
As our second contribution, we put forward a survey-based measure of observable forecast uncertainty shocks (FUS) defined as the difference between cross-sectional stan-dard deviation of ex-post forecast errors and ex-ante forecast disagreement across the analysts in the Bloomberg survey. Many proxies for uncertainty shocks have been used in the literature. For example, Bloom (2009) employs surprise movements in stock mar-ket volatility whereas Jurado, Ludvigson and Ng (2015) use the conditional volatility of forecast errors obtained from a wide range of indicators. Our measure of cross-sectional standard deviation of forecast errors is closest to the one used in Bachmann, Elstner and Sims (2013). FUS turns out to be a weighted version of the squared mean surprise. The weights are time-varying and correspond to the inverse of the sum of disagreement and the standard deviation of forecast errors. In other words, the weights can be seen as the sum of ex-ante and expost “precision” of the forecast. Third, we find evidence of asymmetries in the response of TIPS and IEs to macroeconomic surprises which has not
3The wealth effect plays out as follows: when expected growth is higher, the demand for current
consumption is higher and the demand for borrowing increases which, in turn, leads interest rates to rise.
4Ehling et al. (2016) show that inflation disagreement affect nominal yields through its impact on the
real side of the economy. The sign of the impact also depends on whether the substitution or the income effect dominates.
5Additionally, our study makes use of the Bloomberg survey whereas they use the BlueChip survey. 6Regarding the macroeconomic and financial implications of macroeconomic uncertainty, Segal,
been documented so far.
We expect that the impact of disagreement about economic fundamentals and forecast uncertainty to be different on financial markets since disagreement could simply reflect heterogenous but certain expectations and forecast uncertainty sweeps heterogeneity of the forecasts under the rug. In terms of the sign of the impact, we expect that disagree-ment decreases nominal and real yields and this is line with the findings in Buraschi and Whelan (2016) when we have risk tolerant agents and the substitution effect dominates the income effect.7 On the other hand, we expect forecast uncertainty to increase nominal
and real yields since it very likely increases risk premia. Since FUS is a linear combina-tion of forecast uncertainty and disagreement, the sign of their impact will depend on the latter two and the relative magnitude of each.
The main empirical findings are as follows. Second moments affect nominal yields the most, followed by IEs and TIPS. On the other hand, FUS affect TIPS and IEs the most. Additionally, forecast uncertainty (shocks) about nonfarm payrolls only affect nominal yields significantly and their inclusion strengthens the response of the latter by almost one basis point, across all maturities. Treasury yields react differently to disagreement and forecast uncertainty. This is most likely due to the fact that disagreement com-prises heterogeneity whereas forecast uncertainty does not, by construction. Our findings corroborate the findings in Buraschi and Whelan (2016) regarding disagreement reduc-ing nominal and real yields, if they are not corrected for risk and liquidity premia, as the correction might be introducing noise and blurring the sign. Regarding the sign of the response to forecast uncertainty (shocks), it mainly depends on the macroeconomic announcement, the adjustment of liquidity premia and the prevailing macroeconomic regime. But if we do not correct for liquidity and risk premia, then forecast uncertainty increases nominal and real yields. Furthermore, we find evidence of asymmetric responses to macroeconomic surprises. For nominal yields, sign effects are concentrated in negative surprises whereas for TIPS and IEs, they are concentrated in positive surprises. One potential explanation for this discrepancy is the different sample size for nominal yields versus TIPS and IEs and market specific factors. Moreover, size effects are more present in the response of nominal and IEs rather than in the response of TIPS.
The remainder of this paper is structured as follows. In Section 2.2, the data on survey forecasts, nominal and real yields and market-based IEs are discussed. Additionally, disagreement and forecast uncertainty (shocks) are introduced and analyzed. In Section 2.3, we introduce the event-study framework that we employ and present results for the response of Treasury yields to macroeconomic news, second moments and FUS. In Section 2.4, we study the asymmetric response of Treasury yields to macroeconomic surprises. In Section 2.5, we explore time-variation in the response of financial markets to macroeconomic news and second moments across three macroeconomic regimes and section 2.6 concludes. Appendices 2.A, 2.B and 2.C include Tables with the results, supplementary material and robustness checks.
7Another potential explanation for why disagreement decreases yields goes back to Miller (1977) who
break-even inflation (BEI) rates are derived as the spread between nominal and TIPS of the same maturity. The data is daily and is obtained from the Board of Governor’s web-site8.
BEI rates are not a pure measure of IEs because they also comprise of liquidity and inflation risk premium. We purge liquidity premium from BEI rates9 following the
regression-based method in Strohsal and Winkelmann (2015) and Pflueger and Viceira (2011) using the GARCH standard deviation and AAA spread. In a similar vein, TIPS bonds are also purged from liquidity premium – see Appendix 2.C for details.
Our sample extends from the beginning of 1999 until November 2016 for nominal yields, but only from mid 2004 for TIPS bonds and market-based IEs and that is because only until then that a relatively large number of inflation indexed bonds have been traded in the US secondary market.
2.2.2
Economic news, disagreement, forecast uncertainty and
FUS
The data on macroeconomic releases, the corresponding median forecast, and analysts’ individual forecasts are obtained from the survey conducted by Bloomberg.10 The
sur-vey is conducted as follows: prior to every scheduled macroeconomic release, the sursur-vey participants submit their forecasts for the upcoming release, then these forecasts are published accompanied by the name of the analyst, the institution they are affiliated with and the date of submission of the forecast. Moreover, analysts are allowed to revise their forecasts before the macroeconomic release. Surprises are defined as the difference between the actual economic release, Ait, and the consensus median expectations of the
release, Eit, and then normalized by their standard errors to facilitate comparison across
different news types, as follows: Sit = σ(AitAit−E−Eitit) where i runs from 1 to K, the total
num-ber of macroeconomic news that we consider.
Table 2.1 below reports the economic releases considered, their unit of measurement, the average number of forecasters across the sample and the means and standard de-viations of the corresponding surprises. Although Bloomberg has been conducting this survey since February 1997, the number of participants during the first two years for
8We use zero-coupon US Treasury bonds.
9BEI yields, which we use in this study, contain only liquidity premia whereas forward rates contain
mostly inflation risk premia.
10I am grateful for Jiehui Hu for sharing with me Bloomberg data on announcements, median forecasts,
the majority of economic releases considered is quite low. This can potentially bias our measure of ex-ante uncertainty, therefore we stick to the days for which there are at least 15 forecasters - see de Goeij, Hu and Werker (2013).
Table 2.1: Bloomberg survey forecasts
Data Release Frequency First Release Last Release Units Av. nr. of forecasters Mean Standard deviation
Core CPI Monthly 19/02/1999 17/11/2016 Percent change mom 80.1 -0.004 0.09
GDP (advance) Quarterly 31/07/1998 28/10/2016 Percent change qoq 82.8 -0.034 0.73
Industrial Production Monthly 17/02/1999 16/01/2016 Percent 81.1 -0.047 0.36
Capacity Utilization Monthly 17/02/1999 16/11/2016 Percent 62.37 -0.033 0.33
Non-farm payrolls Monthly 05/03/1999 04/11/2016 Thousands 91.24 -17.91 79.54
Unemployment rate Monthly 05/03/1999 04/11/2016 Percent 85.08 -0.03 0.14
Initial Claims Weekly 25/02/1999 23/11/2016 Thousands 38.57 0.42 17.98
Leading Indicators Monthly 02/03/1999 18/11/2016 Percent 45.09 0.015 0.184
ISM manufacturing Monthly 01/03/1999 01/11/2016 Index 79.53 0.136 1.922
Retail Sales Monthly 13/06/2001 15/11/2016 Percent change mom 82.56 -0.003 0.60
New Home Sales Monthly 02/03/1999 23/11/2016 Thousands 73.45 4.403 59.46
Consumer Confidence Monthly 23/02/1999 29/11/2016 Index 72.20 0.12 4.99
Notes: The average number of forecasters is taken over the sample size of each macroeconomic release. mom denotes month over month and qoq quarter over quarter. Core PPI is not included because individual forecast data was not available at Bloomberg from 2014 onwards.
A relatively large literature is concerned with the measurement of uncertainty and whether disagreement qualifies as a good proxy for ex-ante uncertainty. Disagreement and ex-ante uncertainty are not always equivalent since disagreement also reflects hetero-geneity as highlighted by Zarnowitz and Lambros (1987) and more recently by Bachmann, Elstner and Sims (2013). Notable contributions that use the standard deviation across analysts’ point forecasts, known as disagreement, as a proxy for ex-ante uncertainty in-clude Bomberger (1996), Giordani and Soderlind (2003), Johnson (2004), Bachmann, Elstner and Sims (2013), Andersen et al. (2003), Bloom (2009), among others.
In particular, Bomberger (1996) finds a positive relationship between the dispersion of inflation forecasts from the Livingston survey and the variance of forecast errors as a proxy for uncertainty. Similarly, Giordani and Soderlind (2003)11 find that
disagree-ment about inflation and output growth from Survey of Professional Forecasters (SPF) is positively correlated with uncertainty measures obtained from density forecasts over the sample extending from 1969 to 2001. Moreover, Andersen et al. (2003) proxies un-certainty about economic fundamentals by the standard deviation of expectations across individual forecasters. Using the IFO Business Climate survey, Bachmann, Elstner and Sims (2013) document a strong correlation between ex-post forecast error uncertainty index and ex-ante forecast disagreement in the survey.12
On one hand the Bloomberg survey does not provide us with a density forecast for each individual forecaster, which would be the ideal measure of uncertainty14, unlike the
11Zarnowitz and Lambros (1987) finds a weak positive correlation between disagreement and various
measures of uncertainty using Survey of Professional Forecasters data over a shorter sample.
12Lahiri and Sheng (2010) establish the missing link between disagreement and uncertainty. Using a
decomposition of forecast errors into common and idiosyncratic shocks, they show that the difference between disagreement and the variance of forecast errors,13can be interpreted as the variance of aggregate
shocks that increases with the forecast horizon and depends on the state of the economy. Thus, they conclude that disagreement is not always a good proxy for uncertainty, especially during turbulent economic times and as the forecast horizon increases.
squared forecast errors (ex-post uncertainty) and the standard deviation of individual forecasters’ point forecasts, known as disagreement among forecasters (ex-ante uncer-tainty).
In mathematical terms, the measure of forecast uncertainty shocks takes the following form15: SitF U S≡ SitU − SitD ≡ sPNt j=1(fitj− Ait)2 Nt − sPNt j=1(fitj− ¯fit)2 Nt− 1 (2.1) where fijt is the forecast of forecaster j at time t for fundamental i, ¯fit is the consensus
expectation of each fundamental computed as the sample average across the forecasts available at time t, and j runs from 1 to Nt, the number of forecasters at time t.
Equation 2.1 above is written in terms of time t, although the individual forecasts and the announced values may not take place at the same time since more information is flowing between the date the forecasts are submitted and the release date of funda-mentals. This renders forecasts stale and FUS noisy. This issue is well known in the voluminous literature on macroeconomic news and is usually ignored by assuming that the consensus forecast and the announced value take place at the same time.16 I follow
this convention when constructing uncertainty surprises.
A crucial aspect of event studies is the ability to measure the unexpected component of an event which will, in turn, change the information set of market participants – see Gurkaynak and Wright (2013) for a comprehensive overview. Our measure of uncertainty surprises achieves this because we subtract the expected uncertainty, proxied by disagree-ment among forecasters, before any announcedisagree-ment from ex-post uncertainty that realizes after the value of the fundamental is announced.
In Table 2.2 below we examine whether disagreement is positively correlated with the standard deviation of forecast errors for the macroeconomic announcements in Table 2.1 above. Using the MMS survey17, Gurkaynak and Wolfers (2006) explore whether
disagreement is correlated with the standard deviation of the market-based state price distribution for a small subset of macroeconomic releases (ISM, initial claims, nonfarm payrolls and retail sales) and for the period extending from October, 2002 until July 2005.
15Other measures of disagreement, as the mean absolute deviation used in de Goeij, Hu and Werker
(2013) and the interquartile range in Giordani and Soderlind (2003) can also be used but since we restrict ourselves to days with 15 forecasters, the impact of outliers is already taken care off. Additionally, we use the unbiased estimator of the standard deviation and for that reason we divide by Nt-1 instead of
Nt.
16This problem should not be very pronounced since analysts at the Bloomberg terminal are allowed
to alter their forecast until few hours before the release of macroeconomic variables.
They find that there is strong correlation between the two but the correlation becomes weaker when they focus on lower-frequency variation.
Table 2.2: Disagreement and Uncertainty
Disagreement R2 Core CPI 0.65* (0.24) 0.03 GDP (advance) 1.27** (0.22) 0.27 Industrial Production 1.72** (0.34) 0.38 Capacity Utilization 0.97* (0.38) 0.04 Non-Farm Payrolls 1.31** (0.25) 0.13 Unemployment Rate 1.17** (0.28) 0.08 Initial Claims 1.18** (0.10) 0.35 Leading Indicators 1.40** (0.16) 0.62 ISM manufacturing 0.96* (0.27) 0.05 Retail Sales 1.97** (0.49) 0.49 New Home Sales 1.10** (0.04) 0.79 Consumer Confidence 0.77** (0.18) 0.04
Note: Least square estimation of the standard deviation of forecast errors for macro variables in the first column of the table on disagreement and a constant. HAC standard errors are used. The constant is omitted.**,* denote statistical significance at 5%, 10% level, respectively. Sample sizes as in Table 2.1.
The results in Table 2.2 show that the coefficients on disagreement are statistically significant for all macroeconomic releases considered. For example, disagreement is a very good proxy for the standard deviation of forecast errors in new home sales, leading indicators, retail sales and initial claims because the R2 is relatively high but is a weaker
proxy for core CPI, ISM, capacity utilization and consumer confidence.
Figures 2.1-2.4 plot disagreement, the standard deviation of forecast errors, and FUS for a subset of macroeconomic releases.
.0 .1 .2 .3 .4 .5 .6 .7 2000 2002 2004 2006 2008 2010 2012 2014 2016 ex-ante uncertainty ex-post uncertainty -.05 .00 .05 .10 .15 .20 .25 .30 .35 2000 2002 2004 2006 2008 2010 2012 2014 2016 Uncertainty surprise
0 100 200 2000 2002 2004 2006 2008 2010 2012 2014 2016 ex-ante uncertainty ex-post uncertainty -40 0 40 2000 2002 2004 2006 2008 2010 2012 2014 2016 Uncertainty surprise
Figure 2.2: Uncertainty surprises for New Home Sales
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 2000 2002 2004 2006 2008 2010 2012 2014 2016 ex-ante uncertainty ex-post uncertainty -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 2000 2002 2004 2006 2008 2010 2012 2014 2016 Uncertainty surprise
Figure 2.3: Uncertainty surprises for Capacity utilization
.00 .04 .08 .12 .16 .20 .24 .28 2000 2002 2004 2006 2008 2010 2012 2014 2016 ex-ante uncertainty ex-post uncertainty -.04 .00 .04 .08 .12 .16 .20 2000 2002 2004 2006 2008 2010 2012 2014 2016 Uncertainty surprise
Few stylized facts emerge from observing Figures 2.1-2.4. First, disagreement and ex-post uncertainty co-move, as Table 2.2 above shows, but disagreement typically un-derestimates uncertainty and this is in line with much evidence in the literature see –Lahiri and Sheng (2010), Gurkaynak and Wolfers (2006), Zarnowitz and Lambros (1987), Gior-dani and Soderlind (2003). Second, uncertainty surprises are almost always positive with very few exceptions, usually spike during crisis periods and are stationary, so they do not require any further differencing.
Our uncertainty surprises being almost always nonnegative is most likely due to the fact that the difference between the mean squared forecast errors and the cross-sectional variance of individual forecasts is equal to ( ¯fit−Ait)2 18which is strictly positive. However,
following the vast literature on macroeconomic uncertainty, we deal with the square root of each of these quantities. Additionally, if we scale by Nt instead of Nt−1, SitF U S takes
the following form:
( ¯fit− Ait)2 rP Nt j=1(fitj−Ait)2 Nt + rP Nt j=1(fitj− ¯fit)2 Nt (2.2)
which is a weighted version of the squared mean surprise. The weights are time-varying and correspond to the inverse of the sum of disagreement and the standard deviation of forecast errors. In other words, the weights can be seen as the sum of ex-ante and expost precision of the forecasts.
2.3
Results
Tables that contain the results for all the remaining sections can be found in Appendix A, at the end of this chapter.
2.3.1
The sensitivity of Treasury Yields to Macroeconomic
Sur-prises
The effect of macroeconomic news on asset prices is studied in the following event study, following T. Swanson and C. Williams (2014) and Pericoli and Veronese (2016):
∆yt= β0+ K
X
i=1
βi· Sit+ ǫt (2.3)
where ∆yt corresponds to the daily change, in basis points, of short, medium and long
term nominal yields, TIPS and IEs after a certain announcement, at time t, and ǫt is a
residual representing the influence of other news and other factors on asset prices that day. Moreover, we only include days with at least one macroeconomic release and on days when there is no news about a particular macroeconomic series the corresponding element of Sit is set to zero.19
18This result has been shown in Lahiri and Sheng (2010).
5-year nominal yields by 4.23 basis points which constitutes the largest impact on nomi-nal yields, followed by ISM and retail sales. Although not all the surprises in Table 2.3 are statistically significant at 5 or 10 % level, they are jointly significant as the p-value in the last row of the table suggests. Additionally, these results are broadly in line with the findings in Pericoli and Veronese (2016).20
Table 2.4 shows that TIPS yields respond to a bigger number of surprises compared to nominal yields and that all surprises are jointly significant at the 5% level. For real-side pro-cyclical news and across all maturities, higher-than-expected announcements in nonfarm payrolls, ISM, retail sales and leading indicators21 increase TIPS yields, with
surprises in nonfarm payrolls having the largest effect followed by ISM, retail sales and leading indicators. With respect to counter-cyclical news, not only surprises in initial claims decrease yields but also unemployment news, which was not the case for nominal yields.22 The intuition behind the signs of the responses can be explained as follows: a
higher-than-expected surprise in a procyclical indicator signals that the economy is grow-ing faster than expected and, in turn, the demand for investment goods is expected to increase and the real interest rate would have to rise. Moreover, in contrast to nominal yields, price news (core CPI) decrease TIPS yields and the response is significant even for 10-year yields. The latter sign is intuitive since real interest rates are obtained from nominal interest rates after subtracting IEs. Given that nominal yields, do not increase as a result of CPI news (see Table 2.3), it has to be the case that IEs do; that is indeed the case – see Table 2.5. These results are echoed in other studies as in Bauer (2015) and Zhang (2016), although the sample is not the same.
Regarding the response of IEs, Table 2.5 shows that surprises in nonfarm payrolls, core CPI, capacity utilization, consumer confidence, ISM and new homes increase IEs across all maturities whereas surprises in initial claims and leading indicators23 decrease
them. The negative response of IEs to surprises in leading indicators is due to the fact that TIPS’s reaction is positive (see Table 2.4), while that of nominal yields is not sig-nificantly significant (see Table 2.3). Furthermore, for 10-year IEs, surprises in retail sales and industrial production enter significantly. As was the case for nominal and TIPS yields, IEs react most strongly to nonfarm payrolls announcements.
20They also find that surprises in advance GDP and consumer confidence are statistically significant,
but their sample ends in 2014.
21Surprises in leading indicators enter significantly when using liquidity-adjusted TIPS yields, however,
they fail to be statistically significant if we use unadjusted TIPS yields.
22Unemployment news are statistically significant for 3-year and 5-year TIPS yields only.
23For 10-year IEs, leading indicators are not anymore statistically significant. Moreover, if unadjusted
In summary, all macro releases considered are jointly significant for nominal and TIPS yields and IEs, with the vast majority of the sensitivity being concentrated in TIPS and IEs. Furthermore, while nonfarm payrolls announcements have the largest effect on all yields, the response of nominal yields to the latter is the largest and is almost double the response of either TIPS or IEs.
2.3.2
The Sensitivity of Treasury Yields to Disagreement and
Forecast Uncertainty
In this section, we study the response of nominal yields, TIPS and BEI rates to forecast uncertainty and disagreement by running the following regression:
∆yt = β0+ K X i=1 γi· SitD+ K X i=1 δi· SitU + ǫt (2.4)
The estimates in Table 2.6 show that disagreement about new homes, leading indica-tors, consumer confidence and ISM are individually statistically significant at either 5 or 10 % level whereas only forecast uncertainty about new homes and leading indicators are statistically significant.24 One important observation is that disagreement and forecast
uncertainty, when they enter significantly, affect nominal yields in different ways; while disagreement decreases yields, forecast uncertainty increases them. Additionally, while forecast uncertainty and disagreement about all surprises are jointly significant at 10% level for 3- and 5-year yields, as the p-value in the last row of Table 2.6 shows, they are not anymore for 10-year yields.
Table 2.7 shows that disagreement about capacity utilization and new homes are sta-tistically significant for 3- and 5-year TIPS yields while that for industrial production is only significant for 10-year yields. Moreover, forecast uncertainty about leading indicators is only significant for 3-year maturity. In contrast to the response of nominal yields, dis-agreement about different macroeconomic variables can increase or decrease TIPS yields, while forecast uncertainty always increase yields but it is only significant for short term maturities, unlike disagreement. Moreover, and in line with nominal yields, disagreement and uncertainty are jointly significant only for short and medium term maturities. More-over, the results in Table 2.7 are sensitive to whether liquidity-adjusted TIPS are used or not.25 Therefore, one can draw the following conclusions. First, whether disagreement
increase or decrease TIPS yields depends on the macroeconomic release, for example dis-agreement about capacity utilization always decrease yields while that about industrial
24In the next section, when we include the surprises to the regression, then disagreement and forecast
uncertainty about new variables will appear significant.
25If unadjusted TIPS are used, for 3-year maturity, disagreement about new homes is not anymore
tainty about GDP increase 5-year IEs. In line with nominal and TIPS yields, disagree-ment and uncertainty are jointly statistically significant for short and medium term IEs, as the p-value in the last row of Table 2.8 shows.
In general, the following conclusions can be drawn after analyzing the results in Tables 2.6, 2.7 and 2.8. First, second moments are jointly significant for 3- and 5-year nominal, TIPS and IEs. Second, forecast uncertainty, if significant, increases nominal and TIPS yields while it can increase or decrease IEs, depending whether BEI rates are adjusted for liquidity premia.28 Third, disagreement, if significant, decreases nominal yields,
in-creases IEs, and can increase or decrease TIPS yields, depending on the macroeconomic fundamental considered and whether TIPS are adjusted for liquidity premia or not. In the following section, when we include macroeconomic surprises along with disagreement and forecast uncertainty, we shed led light on the economic channels through which dis-agreement and uncertainty affect nominal yields.
2.3.3
The Sensitivity of Treasury Yields to Macroeconomic
Sur-prises, Forecast Uncertainty and Disagreement
In the previous two sections we studied the impact of macroeconomic surprises, disagree-ment and forecast uncertainty on yields separately. In this section, we study their impact jointly in the following regression:
∆yt = β0+ K X i=1 φi· Sit+ K X i=1 θi· SitD + K X i=1 ρi· SitU + ǫt (2.5)
Table 2.9 presents some interesting and important results. In comparison to Table 2.3, surprises in core CPI are statistically significant at 10% level (only for 3-year yields), in addition to nonfarm payrolls, ISM, retail sales, and initial claims. Moreover, the mag-nitude of the response of nominal yields to nonfarm payrolls surprises has increased by almost one basis point when disagreement and uncertainty are included, and this holds true across all maturities. Another important observation is that forecast uncertainty about nonfarm payrolls enters significantly whereas disagreement does not and the re-verse is true for ISM; while for retail sales and initial claims neither disagreement nor
26When we add the surprises in the next section and we do the analysis for unadjusted TIPS, then
disagreement decreases TIPS.
27If unadjusted IEs are used, disagreement about capacity utilization is not significant anymore while
disagreement about initial claims increases IEs for all maturities and that for retail sales increases 3-year IEs. Moreover, uncertainty about initial claims and GDP do not enter significantly anymore.
28With unadjusted BEI rates, forecast uncertainty always decrease BEI rates, when it enters
forecast uncertainty enter significantly. Disagreement and forecast uncertainty about leading indicators, ISM and new homes are still statistically significant, as in Table 2.6, but that for consumer confidence is not anymore and disagreement about capacity uti-lization becomes statistically significant. Moreover, disagreement and uncertainty about leading indicators are both statistically significant across all maturities. Another robust finding is that even after the inclusion of macroeconomic surprises, disagreement, when it enters significantly, decreases nominal yields while forecast uncertainty increases them. Table 2.10 shows that the surprises that affect TIPS yields significantly across all ma-turities are still the same as in Table 2.4 above, except for capacity utilization which is now significant for 3-year yields at 10% level. In comparison to Table 2.7, two differences emerge: First, disagreement about capacity utilization becomes significant for 10-year yields, in addition to industrial production. Second, uncertainty about ISM decrease 3-year TIPS whereas that about retail sales increase the latter and leading indicators is not anymore statistically significant. Additionally, it seems that disagreement affects TIPS yields significantly across all three maturities whereas uncertainty only affects short term yields. Furthermore, Table 2.10 shows that disagreement and uncertainty can increase or decrease TIPS, however, when unadjusted yields29 are used, we can recover a clear-cut
result as in the case of nominal yields, that disagreement decreases yields whereas uncer-tainty increases them.
Regarding the response of IEs to news, and in comparison to Table 2.5, Table 2.11 shows that surprises in capacity utilization and ISM are not anymore statistically signif-icant for 3- and 5-year IEs, while surprises in GDP become statistically signifsignif-icant at 5% level across all maturities. Additionally, for 10-year IEs, new homes is not statistically significant anymore. In comparison with Table 2.8 and for 3-year IEs, disagreement about capacity utilization is not anymore significant while disagreement about GDP becomes significant for 5-year IEs. Moreover, uncertainty about GDP and unemployment become significant for 3-year IEs, and unemployment for 5-year IEs. As was the case for TIPS, Table 2.11 shows that disagreement and uncertainty can increase or decrease IEs but if unadjusted BEI rates are used30then we obtain a clear-cut result in which disagreement
increase IEs whereas uncertainty decrease them, if significant.
In summary, most of the responsiveness of TIPS and IEs is concentrated in the
sur-29For 3-year TIPS, surprises in unemployment, capacity utilization and new homes are not significant
anymore. Additionally, disagreement about new homes and uncertainty about ISM are not significant anymore. But uncertainty about leading indicators enters significantly and with a positive sign. For 5-year TIPS, surprises in leading indicators and unemployment are not significant anymore. Moreover, leading indicators is the only announcement for which both disagreement and uncertainty are significant across all maturities. For 10-year yields, surprises in unemployment and disagreement about industrial production are not significant anymore.
30For 3-year IEs, surprises in consumer confidence and leading indicators are not significant anymore.
TIPS and IEs depending whether they are adjusted for liquidity premia and on the par-ticular macroeconomic announcement. But, if liquidity-unadjusted TIPS and IEs are used, then we find that disagreement decrease TIPS whereas uncertainty increase them, as was the case for nominal yields. On the other hand, disagreement increases IEs and uncertainty decreases them.
Moreover, the different effect of disagreement and uncertainty likely stems from the fact that disagreement comprises heterogeneity across the analysts whereas uncertainty does not, as Huong (2016) finds. The economic channel through which disagreement leads to the reduction of yields goes back to Buraschi and Whelan (2016), who shows that when the substitution effect overrules the wealth effect, agents consume less, demand on borrowing decrease, and as a result of that interest rates decrease. The most plausible channel through which forecast uncertainty affect yields is by raising risk premia.
2.3.4
The Sensitivity of Treasury Yields to Macroeconomic
Sur-prises and FUS
In this section we study the impact of macroeconomic surprises and FUS on Treasury yields by running the following regression:
∆yt = β0+ K X i=1 γi· Sit+ K X i=1 θi· SitF U S+ ǫt (2.6)
Table 2.12 shows that nominal yields respond significantly to FUS in nonfarm payrolls across all maturities and to that in capacity utilization only for three-year yields. This is in line with the response of nominal yields to forecast uncertainty about nonfarm payrolls (see Table 2.9). Moreover, the inclusion of FUS increases the sensitivity of nominal yields to surprises in nonfarm payrolls by almost one basis point, as was the case when second moments were included in the previous section.
With respect to the response of TIPS yields, Table 2.13 shows that FUS in GDP, ISM, retail sales and leading indicators are statistically significant31 for three- and
five-year maturities while only FUS in GDP and leading indicators remain significant for long-term yields32. More importantly, FUS can increase or decrease TIPS yields
depend-ing on the macroeconomic announcement, for example, FUS in GDP and ISM decrease TIPS yields whereas that in retail sales and leading indicators increase them. It is also
31In the previous section, forecast uncertainty about ISM and GDP are statistically significant. 32If unadjusted TIPS are used, FUS in new homes become significant for three-year yields and FUS in
possibly related to which effect dominates (i.e., positive or negative effect of disagreement or positive effect of uncertainty). Additionally, this conclusion remains valid even when we do not adjust TIPS for liquidity premium.
Table 2.14 shows that IEs respond significantly to FUS in GDP and retail sales across all maturities and to that in unemployment, initial claims and consumer confidence for three and five-year IEs33. Those results are very similar to the ones obtained in the
pre-vious section when we examined the impact of forecast uncertainty on IEs. Moreover, the opposite responses of IEs and TIPS to FUS in GDP and retail sales can be reconciled as follows: IEs are obtained from subtracting TIPS from nominal yields34, therefore, the
sign of the response of IEs should be opposite to that of TIPS, given that nominal yields do not react significantly to the aforementioned announcements.
In summary, FUS in macroeconomic announcements increase nominal yields but they tend to increase or decrease TIPS yields and IEs depending on the particular macroeco-nomic announcement and whether the effect of disagreement or uncertainty dominates, regardless whether they are corrected for liquidity premia. In contrast to the previous section, most of the responsiveness of TIPS and IEs, is not only concentrated in the surprises themselves but also in FUS, while nominal yields respond significantly only to FUS in nonfarm payrolls.
2.4
Asymmetries in the response of Treasury Yields
to Surprises
In this section, we study whether the response of nominal, TIPS and IEs is asymmetric due to the sign and size of surprises. Important contributions include Andersen et al. (2003) and Ehrmann and Fratzscher (2005) who find that negative and large surprises have a stronger impact on exchange rates. For the bond market, Hautsch and Hess (2007) document that bad news about nonfarm payrolls have a stronger impact on bond prices compared to good news. To check whether the size and sign of surprises matter, we estimate the following regression:
∆yt= β0+ K X i=1 (1Sit>0β1iSit+1Sit<0β2iSit+ β3iS 2 it) + ǫt (2.7) where S2
it is the squared surprise35 and 1Sit>0,1Sit<0 are indicator functions taking values
1 when Sit is bigger or less than 0, respectively; both dummies being zero otherwise.
33If unadjusted IEs are used, FUS in unemployment and consumer confidence fail to be significant for
three-year yields. For five-year yields, FUS in initial claims and unemployment also fail to be statistically significant. Additionally, FUS in GDP is not significant for ten-year yields - see Table 2.27 in the Appendix 2.B.
34This is only true if liquidity premia are accounted for, as we do in our analysis, otherwise, IEs should
be replaced by break-even inflation rates.
35To make the estimates comparable, we standardize the squared surprise by the sample standard
significantly but the magnitude of negative surprises is larger than that of positive sur-prises across all maturities. With respect to new homes, positive sursur-prises are statistically significant instead of negative surprises and for consumer confidence both positive and negative surprises are statistically significant but they enter with different signs. Regard-ing size effects, large surprises in GDP, ISM, nonfarm payrolls and consumer confidence increase nominal yields, which is in line with the sign of the macroeconomic surprise, whereas large surprises in new homes decrease them. This is evidence that there is a nonlinearity in the response to surprises in new homes: small surprises increase yields, while large surprises decrease them.
In comparison to nominal yields, Table 2.16 provides little evidence of size and sign effects for TIPS yields. Sign effects are present for retail sales, ISM, GDP, leading indi-cators and core CPI. In particular, negative surprises in ISM, retail sales and GDP are statistically significant while positive surprises in GDP, core CPI and leading indicators are significant. Additionally for five- and ten-year maturities, positive surprises in retail sales are significant while their negative counterparts are not anymore. In general, the sign effects for TIPS are concentrated mostly in positive rather than negative surprises, especially for medium and long term maturities. Size effects are not strong, as only large surprises in industrial production and GDP enter significantly.
With respect to the asymmetric response of IEs, Table 2.17 shows, as in the case of TIPS, that sign effects are concentrated in positive rather than negative surprises and that is true across all maturities. For example, positive surprises in core CPI, ISM, non-farm payrolls and leading indicators are statistically significant while negative surprises in nonfarm payrolls, consumer confidence and GDP are only significant for five and ten-year maturities. In contrast to TIPS and more in line with nominal yields, size effects are present but vary across maturities. In particular, large surprises in core CPI, retail sales, new homes, GDP, ISM and leading indicators are statistically significant.
2.4.1
FUS and Squared surprises
In this section, we investigate whether our measure of FUS has different implications than if we used squared macroeconomic surprises as regressors. To achieve the latter, we run the following regression where we replace squared surprises by FUS:
∆yt= β0+ K X i=1 (1Sit>0β1iSit+1Sit<0β2iSit+ β3iS F U S it ) + ǫt (2.8)
Regarding the response of nominal yields and comparing Tables 2.15 and 2.18, we notice that FUS and squared surprises are statistically significant for different macroe-conomic announcements. For example, FUS in unemployment and industrial production are statistically significant (see Table 2.15) while squared surprises in these announce-ments are not. Moreover, squared surprises in GDP, nonfarm payrolls and new homes are significant (see Table 2.12) while FUS in these announcements are not. On the other hand, FUS and squared surprises in ISM are both statistically significant, at least for three-year yields.
In a similar vein, comparing Tables 2.16 and 2.19 for TIPS yields, we find that FUS in core CPI and unemployment are statistically significant, while squared surprises in industrial production and GDP are significant. Regarding the response of IEs, Tables 2.17 and 2.20 show that FUS and squared surprises in core CPI and GDP are statisti-cally significant. Moreover, squared surprises in new homes, retail sales, and ISM are significant while FUS in the aforementioned announcements are not.
FUS and squared surprises are not only mathematically different but also economically different because the response of Treasury yields, as we illustrated above, is different and it primarily depends on the macroeconomic announcement.
2.5
Time variation in the response of Treasury yields
In this section we explore time variation in the response of three-year nominal, TIPS and IEs to macroeconomic surprises and second moments, as in equation 2.5 in Section 2.3.3. We split our sample into three periods, following Pericoli and Veronese (2016). The three periods are as follows: a tranquil period extending from March 199936 to July 2007, a
crisis period from August 2007 to March 2009, and a zero lower bound or unconventional monetary policy period from April 2009 till the end of our sample.
Table 2.21 shows that the sensitivity of yields to macroeconomic surprises and second moments varies across the macroeconomic regimes. For the tranquil regime and in com-parison to the full sample results, the sensitivity of yields to surprises in initial claims, ISM and nonfarm payrolls has strengthened while core CPI is not anymore significant. As for disagreement, only capacity utilization remains significant in the tranquil regime while that about ISM, leading indicators and new homes are not significant anymore.
36The starting date of this macroeconomic regime applies to nominal yields. For real yields and BEI
QE regime, we find that the response, in absolute values, to most of the macroeconomic surprises has decreased compared to the tranquil period, while only the response to retail sales increased. The reduction in the sensitivity is due to the zero lower bound (ZLB), as documented by T. Swanson and C. Williams (2014). Moreover, disagreement and uncer-tainty about macroeconomic announcements are rarely significant, with one exception, disagreement about capacity utilization.
In summary, there is evidence of time variation in the response of nominal yields to news and second moments. In particular, the sensitivity to the latter intensifies during the crisis regime while that of the former diminishes in the regime where the ZLB binds. Table 2.22 presents estimates of the sensitivity of TIPS yields to macroeconomic sur-prises and second moments. In comparison to the tranquil regime and in contrast to nominal yields, most of the sensitivity to macroeconomic surprises does not disappear in the wake of the financial crisis. In particular, surprises in GDP, initial claims, and new homes become significant while that in nonfarm payrolls and unemployment fail to be statistically significant. On the other hand, disagreement and uncertainty about no macroeconomic announcement is statistically significant during the crisis. During the QE regime, we observe that surprises in CPI and leading indicators are now significant which was not the case during the tranquil and crisis regimes, in addition to initial claims, ISM and nonfarm payrolls. Additionally, disagreement about new homes and uncertainty about initial claims and nonfarm payrolls become significant. This shows that the finan-cial crisis and the ZLB did not weaken the response of TIPS yields to macroeconomic surprises and this is broadly in line with the findings of Zhang (2016), in a sense that TIPS yields do not become less sensitive to news at the ZLB, in contrast to nominal yields. With respect to the response of IEs, Table 2.23 shows that in contrast to the full sample results, disagreement about GDP, capacity utilization and industrial production are statistically significant during the tranquil period. However, during the financial cri-sis, the responsiveness to the majority of the surprises and second moments disappears, except for nonfarm payrolls news. This suggests that during the crisis regime the con-nection between economic fundamentals and financial markets is muted because of the panic and chaos that the financial crisis caused and the ZLB which became binding since December 2008. Although the sensitivity to macroeconomic news during the QE regime goes back to normal, the response to second moments is rather weak.
weak-ened the connection between economic fundamentals and financial markets. However, the sensitivity of TIPS yields to macroeconomic surprises was not weakened by the crisis and the presence of ZLB, as Zhang (2016) suggests as well.
2.6
Conclusion
In this paper we study the impact of macroeconomic surprises, disagreement and forecast uncertainty on short, medium and long-term nominal, TIPS and market-based IEs for the period extending from 1999 until 2016. Relative to the extant literature, we make three contributions. First, we study the first moment response of Treasury yields to second moments (i.e., disagreement and forecast uncertainty), which has not been studied so far. Second, we put forward a survey-based measure of observable forecast uncertainty shocks defined as the difference between cross-sectional standard deviation of forecast errors and forecast disagreement across the analysts in the Bloomberg survey. Third, we find evidence of asymmetries in the response of TIPS and IEs to macroeconomic surprises which has not been documented before.
Three-year Five-year Ten-year ˆ
βi βiˆ βiˆ
Core CPI 0.89 0.74 0.49 (0.54) (0.60) (0.64) GDP (advance) -0.05 0.05 0.13 (2.43) (2.25) (1.85) Initial Claims -1.22** -1.20** -1.03** (0.28) (0.30) (0.32) ISM manufacturing 3.27** 3.39** 3.17** (0.54) (0.53) (0.52) Nonfarm Payrolls 4.38** 4.23** 3.59** (0.66) (0.68) (0.65) Unemployment -0.57 -0.38 -0.08 (0.50) (0.52) (0.54) Capacity utilization 0.77 (0.73) 0.61 (0.82) (0.83) (0.80) Industrial Production 0.97 0.82 0.50 (0.67) (0.74) (0.81) Retail Sales 2.93** 2.99** 2.73** (0.71) (0.79) (0.87) Consumer Confidence 0.53 0.49 0.39 (0.57) (0.62) (0.65) Leading Indicators 0.48 0.58 0.72 (0.38) (0.46) (0.57) New homes 0.29 0.24 0.16 (0.55) (0.56) (0.54) Observations 2188 2190 2192 R2 0.067 0.059 0.043 H0: β = 0 (p-value) 0.000 0.000 0.000
Notes: ˆβipertain to OLS estimates of regression (2.3). The dependent variable is the daily change, in basis points, of nominal yields. ** denotes significance at 5% level and * at 10% level. HAC standard errors with Bartlett kernel are used and shown in paranthesis. Sample
period: October 30, 1998 to November 29, 2016. H0: β = 0 (p-value)
Table 2.4: Regressions of daily TIPS yield changes on macroeconomic surprises
TIPS Yields
Three-year Five year Ten-year
ˆ
βi βiˆ βiˆ
Core CPI -2.02** -1.51** -1.00** (0.6) (0.57) (0.49) GDP (advance) 0.67 1.01 0.62 (0.91) (0.89) (0.84) Initial Claims -0.95** -0.86** -0.66** (0.25) (0.24) (0.23) ISM manufactuirng 2.34** 2.21** 1.99** (0.61) (0.45) (0.38) Nonfarm Payrolls 2.55** 2.80** 2.61** (0.89) (0.83) (0.63) Unemployment -1.15* -1.05* -0.79 (0.63) (0.58) (0.49) Capacity utilization 0.59 0.54 0.31 (0.71) (0.66) (0.60) Industrial Production -0.45 0.50 0.60 (1.42) (0.62) (0.47) Retail Sales 1.61** 1.89** 1.47** (0.40) (0.34) (0.37) Consumer Confidence -0.28 -0.45 -0.15 (0.61) (0.62) (0.54) Leading Indicators 0.97** 0.93** 1.13** (0.42) (0.42) (0.39) New homes 0.45 0.42 0.24 (0.40) (0.38) (0.32) Observations 1512 1512 1512 R2 0.046 0.054 0.049 H0: β = 0 (p-value) 0.000 0.000 0.000
Notes: ˆβipertain to OLS estimates of regression (2.3). The dependent variable is the daily change, in basis points, of liquidity-adjusted TIPS yields. ** denotes significance at 5% level and * at 10% level. HAC standard errors with Bartlett kernel are used and shown in
paranthe-sis. Sample period: June 1, 2004 to November 29, 2016. H0: β = 0
Table 2.5: Regressions of daily IEs changes on macroeconomic surprises
IEs
Three-year Five year Ten-year
ˆ
βi βiˆ βiˆ
Core CPI 2.44** 1.50** 1.08** (0.57) (0.56) (0.37) GDP (advance) 0.81 0.61 0.59 (0.96) (1.01) (0.53) Initial Claims -0.39* -0.57** -0.63** (0.21) (0.18) (0.18) ISM manufactuirng 0.64* 0.58 0.54 (0.37) (0.52) (0.33) Nonfarm Payrolls 2.59** 2.12** 1.28** (0.54) (0.37) (0.30) Unemployment 0.26 0.34 0.15 (0.49) (0.40) (0.30) Capacity utilization 1.87** 1.58** 1.06** (0.80) (0.77) (0.45) Industrial Production -0.91 -1.39 -1.43** (1.28) (1.04) (0.53) Retail Sales 0.90 0.82 1.06** (0.68) (0.60) (0.34) Consumer Confidence 0.86* 0.93* 0.65** (0.48) (0.48) (0.29) Leading Indicators -1.13** -0.90** -0.43 (0.51) (0.42) (0.51) New homes 0.25 0.24 0.38* (0.36) (0.34) (0.22) Observations 1514 1514 1514 R2 0.034 0.033 0.047 H0: β = 0 (p-value) 0.000 0.000 0.000
Notes: ˆβipertain to OLS estimates of regression (2.3). The dependent variable is the daily change, in basis points, of liquidity-adjusted BEI rates. ** denotes significance at 5% level and * at 10% level. HAC standard errors with Bartlett kernel are used and shown in
paranthe-sis. Sample period: June 1, 2004 to November 29, 2016. H0: β = 0
Table 2.6: Regressions of daily nominal yield changes on disagreement and forecast un-certainty
Nominal Yields
Three-year Five-year Ten-year
ˆ
γi ˆδi ˆγi δiˆ ˆγi δiˆ
Core CPI 0.04 -0.22 0.02 -0.26 -0.53 -0.29 (0.31) (0.63) (0.33) (0.67) (0.41) (0.69) GDP (advance) -0.22 -0.92 -0.44 -0.60 -0.66 -0.16 (1.42) (2.67) (1.37) (2.48) (1.22) (2.06) Initial Claims 0.56 -0.56 0.61 -0.58 0.59 -0.55 (0.37) (0.36) (0.42) (0.39) (0.45) (0.42) ISM manufactuirng -0.71* 0.79 -0.66* 0.80 -0.50 0.71 (0.36) (0.77) (0.36) (0.75) (0.34) (0.69) Nonfarm Payrolls 0.13 -0.58 0.04 -0.66 -0.04 -0.69 (0.62) (0.84) (0.63) (0.80) (0.61) (0.69) Unemployment 0.31 -0.21 0.49 -0.31 0.66 -0.45 (0.59) (0.57) (0.59) (0.58) (0.54) (0.57) Capacity utilization -0.60 0.35 -0.73 0.39 -0.82 0.43 (0.43) (0.59) (0.49) (0.66) (0.54) (0.74) Industrial Production 1.04 -1.32 0.97 -1.18 0.77 -0.83 (0.73) (0.84) (0.76) (0.83) (0.78) (0.80) Retail Sales -0.78 1.16 -0.83 0.93 -0.78 0.52 (0.74) (1.36) (0.80) (1.47) (0.84) (1.48) Consumer Confidence -0.59* 0.36 -0.61 0.30 -0.53 0.18 (0.35) (0.51) (0.38) (0.57) (0.41) (0.63) Leading Indicators -1.56* 1.62* -1.84** 1.91** -2.02** 2.08** (0.92) (0.92) (0.80) (0.84) (0.83) 0.90 New homes -3.09** 1.97* -2.89** 1.84 -2.39* 1.55 (1.29) (1.10) (1.36) (1.12) (1.35) (1.09) Observations 2188 2190 2192 R2 0.016 0.015 0.012 H0: β = 0 (p-value) 0.07 0.09 0.17
Notes: ˆγiand ˆδipertain to OLS estimates of regression (2.4). The dependent variable is the daily change, in basis points, of nominal yields. ** denotes significance at 5% level and * at 10% level. HAC standard errors with Bartlett kernel are used and shown
in parentheses. Sample period: October 30, 1998 to November 29, 2016. H0: β = 0
Table 2.7: Regressions of daily TIPS yield changes on disagreement and forecast uncer-tainty
TIPS Yields
Three-year Five year Ten-year
ˆ
γi ˆδi ˆγi δiˆ γiˆ δiˆ
Core CPI -0.36 0.40 -0.25 0.09 -0.30 0.16 (0.31) (0.50) (0.26) (0.43) (0.22) (0.34) GDP (advance) -0.67 -0.51 -0.55 -0.46 -0.36 -0.25 (0.53) (0.73) (0.43) (0.64) (0.53) (0.75) Initial Claims -0.10 0.008 0.06 0.007 -0.09 0.04 (0.25) (0.06) (0.21) (0.05) (0.24) (0.06) ISM manufactuirng -0.02 -0.70 -0.01 -0.29 -0.12 0.09 (0.43) (0.87) (0.36) (0.72) (0.32) (0.64) Nonfarm Payrolls 0.20 0.37 0.07 0.12 0.13 -0.21 (0.88) (1.10) (0.76) (1.02) (0.59) (0.73) Unemployment 0.13 -0.67 0.42 -0.58 0.52 -0.49 (0.64) (0.71) (0.65) (0.69) (0.54) (0.58) Capacity utilization -1.29* 0.98 -0.75* 0.27 -0.56 -0.10 (0.68) (1.20) (0.43) (0.64) (0.37) (0.52) Industrial Production 0.69 0.38 1.13 -0.61 1.16** -0.81 (1.03) (1.53) (0.72) (0.87) (0.58) (0.61) Retail Sales -0.15 0.59 0.12 0.22 0.15 0.01 (0.46) (0.53) (0.44) (0.53) (0.35) (0.45) Consumer Confidence 0.72 -0.94 0.52 -0.69 0.56 -0.84 (0.73) (0.94) (0.65) (0.87) (0.53) (0.71) Leading Indicators -0.85 1.36* -0.54 0.99 -0.31 0.77 (0.71) (0.75) (0.63) (0.68) (0.51) (0.56) New homes 0.70* -0.23 0.67* -0.23 0.37 -0.22 (0.42) (0.51) (0.39) (0.50) (0.35) (0.43) Observations 1512 1512 1512 R2 0.031 0.02 0.017 H0: β = 0 (p-value) 0.002 0.040 0.530
Table 2.8: Regressions of daily IEs changes on disagreement and forecast uncertainty
IEs
Three-year Five year Ten-year
ˆ
γi δiˆ γiˆ δiˆ ˆγi ˆδi
Core CPI 0.64 -1.06 0.42 -0.65 0.29 -0.57 (0.48) (0.73) (0.40) (0.65) (0.22) (0.42) GDP (advance) -0.49 1.36 -0.73 1.59* -0.15 0.47 (0.68) (0.99) (0.52) (0.83) (0.37) (0.54) Initial Claims 0.54 -0.12* 0.33 -0.10** 0.41** -0.07 (0.38) (0.06) (0.27) (0.05) (0.18) (0.05) ISM manufactuirng -0.11 0.36 -0.04 0.09 0.13 -0.04 (0.27) (0.43) (0.27) (0.53) (0.17) (0.32) Nonfarm Payrolls -0.09 -0.47 0.08 -0.76 0.02 -0.39 (0.74) (0.79) (0.59) (0.51) (0.56) (0.32) Unemployment 0.04 0.76 0.12 0.47 0.16 0.20 (0.52) (0.47) (0.46) (0.37) (0.41) (0.28) Capacity utilization 1.04* -1.75 0.64 -0.92 0.10 -0.20 (0.58) (1.10) (0.51) (0.92) (0.27) (0.57) Industrial Production -0.58 0.51 -0.58 0.59 -0.71 0.97 (1.03) (1.71) (0.93) (1.53) (0.56) (0.86) Retail Sales 0.35 -1.41** -0.05 -0.94* -0.03 -0.44 (0.37) (0.43) (0.44) (0.48) (0.35) (0.39) Consumer Confidence -0.49 0.35 -0.23 0.02 -0.30 0.34 (0.61) (0.77) (0.44) (0.60) (0.29) (0.41) Leading Indicators 1.44 -1.48 0.91 -1.05 -0.35 -0.02 (1.88) (1.78) (1.22) (1.17) (0.42) (0.47) New homes -0.21 0.02 0.15 -0.33 0.18 -0.03 (0.37) (0.39) (0.39) (0.37) (0.21) (0.26) Observations 1514 1514 1514 R2 0.025 0.022 0.023 H0: β = 0 (p-value) 0.021 0.026 0.35
Table 2.9: Regressions of daily nominal yield changes on macroeconomic news, disagree-ment and forecast uncertainty
Nominal Yields
Three-year Five year Ten-year
φi θi ρi φi θi ρi φi θi ρi
Core CPI 0.96* 0.04 -0.24 0.82 0.02 -0.27 0.56 −0.00 -0.30 (0.54) (0.28) (0.56) (0.60) (0.31) (0.61) (0.64) (0.33) (0.64) GDP (advance) -0.13 -0.04 -1.125 0.006 -0.28 -0.79 0.14 -0.54 -0.30 (2.66) (1.62) (2.96) (2.46) (1.55) (2.74) (2.02) (1.35) (2.26) Initial Claims -1.19** 0.56 -0.47 -1.18** 0.61 -0.50 -1.01** 0.58 -0.48 (0.27) (0.39) (0.34) (0.29) (0.44) (0.37) (0.31) (0.46) (0.39) ISM manufacturing 3.34** -0.56* 0.37 3.43** -0.50* 0.37 3.17** -0.36 0.31 (0.53) (0.30) (0.55) (0.51) (0.29) (0.53) (0.50) (0.29) (0.51) Nonfarm Payrolls 5.32** -0.01 1.56** 5.11** -0.09 1.40** 4.29** -0.16 1.04* (0.64) (0.46) (0.67) (0.62) (0.50) (0.64) (0.60) (0.53) (0.61) Unemployment -0.54 -0.18 -0.35 -0.34 -0.005 -0.39 -0.03 0.21 -0.44 (0.50) (0.35) (0.54) (0.53) (0.37) (0.57) (0.56) (0.39) (0.59) Capacity utilization 1.18 -0.85* 0.70 1.18 -0.98* 0.71 1.06 -1.03* 0.67 (0.88) (0.47) (0.66) (0.89) (0.53) (0.73) (0.87) (0.57) (0.79) Industrial Production 0.62 0.91 -0.88 0.43 -0.80 -0.80 0.11 0.74 -0.59 (0.74) (0.73) (0.78) (0.82) (0.78) (0.79) (0.89) (0.80) (0.78) Retail Sales 2.93** -0.37 0.29 3.10** -0.41 0.02 2.96** -0.39 -0.34 (0.73) (0.61) (0.87) (0.79) (0.65) (0.95) (0.85) (0.70) (1.03) Consumer Confidence 0.45 -0.56 0.34 0.42 -0.59 0.28 0.33 -0.51 0.16 (0.58) (0.35) (0.51) (0.63) (0.38) (0.57) (0.66) (0.41) (0.63) Leading Indicators 0.37 -1.48* 1.52* 0.47 -1.75** 1.77** 0.61 -1.91** 1.90** (0.38) (0.89) (0.89) (0.46) (0.79) (0.82) (0.57) (0.84) (0.91) New homes 0.03 -3.08** 1.94 0.006 -2.90* 1.82 -0.03 -2.43 1.56 (0.65) (1.48) (1.29) (0.67) (1.56) (1.32) (0.65) (1.53) (1.28) Observations 2188 2190 2192 R2 0.087 0.077 0.058 H0: β = 0 (p-value) 0.000 0.000 0.000
Table 2.10: Regressions of daily TIPS yield changes on macroeconomic news, disagree-ment and forecast uncertainty
TIPS Yields
Three-year Five year Ten-year
ˆ
φi θiˆ ρiˆ φiˆ θiˆ ρiˆ φiˆ θiˆ ρiˆ
Core CPI -1.90** -0.36 0.33 -1.55** -0.28 0.06 -1.10** -0.31 0.14 (0.59) (0.29) (0.49) (0.61) (0.26) (0.43) (0.54) (0.22) (0.36) GDP (advance) 0.20 -0.58 -0.56 0.64 -0.57 -0.31 0.38 -0.35 -0.19 (0.76) (0.59) (0.85) (0.92) (0.50) (0.81) (0.89) (0.58) (0.83) Initial Claims -0.90** -0.12 0.02 -0.85** 0.05 0.02 -0.66** -0.10 0.06 (0.26) (0.24) (0.06) (0.25) (0.22) (0.06) (0.24) (0.24) (0.06) ISM manufactuirng 2.48** 0.04 -0.85* 2.27** 0.04 -0.43 1.99** -0.07 -0.023 (0.55) (0.31) (0.50) (0.43) (0.26) (0.38) (0.38) (0.24) (0.36) Nonfarm Payrolls 2.75** 0.29 0.74 2.96** 0.29 0.55 2.72** 0.20 0.19 (0.78) (0.78) (0.94) (0.71) (0.46) (0.84) (0.55) (0.48) (0.60) Unemployment -1.28** 0.03 -0.94 -1.15* 0.29 -0.81 -0.86* 0.39 -0.65 (0.75) (0.49) (0.78) (0.66) (0.46) (0.72) (0.51) (0.37) (0.58) Capacity utilization 1.83* -1.49** 1.20 1.28 -0.91** 0.45 0.76 -0.65* -0.006 (0.97) (0.68) (1.22) (0.78) (0.41) (0.68) (0.66) (0.36) (0.57) Industrial Production -0.86 0.91 0.26 0.10 1.16 -0.42 0.24 1.14* -0.63 (1.28) (0.93) (1.26) (0.74) (0.73) (0.81) (0.60) (0.60) (0.61) Retail Sales 1.71** -0.21 0.74* 1.95** 0.07 0.38 1.50** 0.11 0.13 (0.36) (0.41) (0.42) (0.35) (0.38) (0.40) (0.40) (0.32) (0.43) Consumer Confidence -0.28 0.72 -0.93 -0.45 0.52 -0.67 -0.12 0.57 -0.82 (0.63) (0.73) (0.94) (0.64) (0.65) (0.87) (0.56) (0.53) (0.71) Leading Indicators 0.82* -0.59 1.08 0.83* -0.29 0.71 1.04** -0.05 0.45 (0.44) (0.69) (0.72) (0.42) (0.61) (0.63) (0.38) (0.48) (0.51) New homes 0.47 0.80* -0.35 0.47 0.77* -0.35 0.31 0.45 -0.31 (0.45) (0.44) (0.56) (0.41) (0.05) (0.53) (0.34) (0.37) (0.45) Observations 1512 1512 1512 R2 0.080 0.070 0.06 H0: β = 0 (p-value) 0.000 0.000 0.000
Table 2.11: Regressions of daily IEs changes on macroeconomic news, disagreement and forecast uncertainty
IEs
Three-year Five year Ten-year
ˆ
φi θiˆ ρiˆ φiˆ θiˆ ρiˆ φiˆ θiˆ ρiˆ
Core CPI 2.31** 0.53 -0.72 1.45** 0.35 -0.43 1.12** 0.25 -0.41 (0.53) (0.46) (0.66) (0.51) (0.39) (0.60) (0.36) (0.19) (0.34) GDP (advance) 1.55** -0.77 1.98** 1.41* -0.96* 2.12** 0.93** -0.27 0.78 (0.61) (0.68) (0.95) (0.81) (0.55) (0.89) (0.42) (0.40) (0.58) Initial Claims -0.40** 0.58 -0.12* -0.55** 0.37 -0.10** -0.62** 0.44** -0.07 (0.20) (0.36) (0.06) (0.17) (0.26) (0.04) (0.18) (0.16) (0.04) ISM manufactuirng 0.57 -0.08 0.33 0.55 -0.01 0.06 0.50 0.16 -0.08 (0.39) (0.26) (0.41) (0.55) (0.26) (0.52) (0.33) (0.16) (0.32) Nonfarm Payrolls 2.66** -0.12 0.07 2.11** 0.05 -0.32 1.31** 0.013 -0.12 (0.55) (0.59) (0.57) (0.38) (0.47) (0.38) (0.30) (0.50) (0.30) Unemployment 0.64 -0.22 0.99** 0.58 -0.09 0.67* 0.29 0.04 0.31 (0.52) (0.39) (0.47) (0.41) (0.35) (0.36) (0.30) (0.34) (0.27) Capacity utilization 1.05 0.66 -1.38 1.17 0.34 -0.61 1.03** -0.13 0.05 (0.96) (0.65) (1.21) (0.90) (0.57) (1.02) (0.51) (0.28) (0.60) Industrial Production -0.41 -0.46 0.64 -1.13 -0.34 0.41 -1.30** -0.45 0.68 (0.99) (1.01) (1.53) (0.81) (0.93) (1.39) (0.43) (0.56) (0.78) Retail Sales 0.74 0.38 -1.36** 0.71 -0.05 -0.87* 1.01** -0.06 -0.35 (0.54) (0.38) (0.48) (0.46) (0.43) (0.51) (0.29) (0.32) (0.33) Consumer Confidence 0.89** -0.47 0.32 0.95* -0.21 -0.01 0.63** -0.29 0.32 (0.48) (0.60) (0.75) (0.49) (0.44) (0.59) (0.28) (0.29) (0.39) Leading Indicators -1.09** 1.21 -1.17 -0.85* 0.74 -0.82 -0.36 -0.42 0.08 (0.45) (1.90) (1.80) (0.44) (1.23) (1.18) (0.50) (0.41) (0.45) New homes 0.31 -0.12 -0.13 0.35 0.24 -0.49 0.43 0.28 -0.21 (0.38) (0.38) (0.46) (0.35) (0.41) (0.44) (0.26) (0.21) (0.29) Observations 1514 1514 1514 R2 0.057 0.053 0.069 H0: β = 0 (p-value) 0.000 0.000 0.000
